sys/netinet/ip_output.c

   1 /*
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  30  * $FreeBSD: src/sys/netinet/ip_output.c,v 1.99.2.37 2003/04/15 06:44:45 silby Exp $
  31  * $DragonFly: src/sys/netinet/ip_output.c,v 1.41 2008/01/06 16:55:52 swildner Exp $
  32  */
  33
  34 #define _IP_VHL
  35
  36 #include "opt_ipfw.h"
  37 #include "opt_ipdn.h"
  38 #include "opt_ipdivert.h"
  39 #include "opt_ipfilter.h"
  40 #include "opt_ipsec.h"
  41 #include "opt_mbuf_stress_test.h"
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/kernel.h>
  46 #include <sys/malloc.h>
  47 #include <sys/mbuf.h>
  48 #include <sys/protosw.h>
  49 #include <sys/socket.h>
  50 #include <sys/socketvar.h>
  51 #include <sys/proc.h>
  52 #include <sys/sysctl.h>
  53 #include <sys/thread2.h>
  54 #include <sys/in_cksum.h>
  55
  56 #include <net/if.h>
  57 #include <net/netisr.h>
  58 #include <net/pfil.h>
  59 #include <net/route.h>
  60
  61 #include <netinet/in.h>
  62 #include <netinet/in_systm.h>
  63 #include <netinet/ip.h>
  64 #include <netinet/in_pcb.h>
  65 #include <netinet/in_var.h>
  66 #include <netinet/ip_var.h>
  67
  68 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options");
  69
  70 #ifdef IPSEC
  71 #include <netinet6/ipsec.h>
  72 #include <netproto/key/key.h>
  73 #ifdef IPSEC_DEBUG
  74 #include <netproto/key/key_debug.h>
  75 #else
  76 #define KEYDEBUG(lev,arg)
  77 #endif
  78 #endif /*IPSEC*/
  79
  80 #ifdef FAST_IPSEC
  81 #include <netproto/ipsec/ipsec.h>
  82 #include <netproto/ipsec/xform.h>
  83 #include <netproto/ipsec/key.h>
  84 #endif /*FAST_IPSEC*/
  85
  86 #include <net/ipfw/ip_fw.h>
  87 #include <net/dummynet/ip_dummynet.h>
  88
  89 #define print_ip(x, a, y)        kprintf("%s %d.%d.%d.%d%s",\
  90                                 x, (ntohl(a.s_addr)>>24)&0xFF,\
  91                                   (ntohl(a.s_addr)>>16)&0xFF,\
  92                                   (ntohl(a.s_addr)>>8)&0xFF,\
  93                                   (ntohl(a.s_addr))&0xFF, y);
  94
  95 u_short ip_id;
  96
  97 #ifdef MBUF_STRESS_TEST
  98 int mbuf_frag_size = 0;
  99 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 100         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 101 #endif
 102
 103 static struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *);
 104 static struct ifnet *ip_multicast_if(struct in_addr *, int *);
 105 static void     ip_mloopback
 106         (struct ifnet *, struct mbuf *, struct sockaddr_in *, int);
 107 static int      ip_getmoptions
 108         (struct sockopt *, struct ip_moptions *);
 109 static int      ip_pcbopts(int, struct mbuf **, struct mbuf *);
 110 static int      ip_setmoptions
 111         (struct sockopt *, struct ip_moptions **);
 112
 113 int     ip_optcopy(struct ip *, struct ip *);
 114
 115
 116 extern  struct protosw inetsw[];
 117
 118 /*
 119  * IP output.  The packet in mbuf chain m contains a skeletal IP
 120  * header (with len, off, ttl, proto, tos, src, dst).
 121  * The mbuf chain containing the packet will be freed.
 122  * The mbuf opt, if present, will not be freed.
 123  */
 124 int
 125 ip_output(struct mbuf *m0, struct mbuf *opt, struct route *ro,
 126           int flags, struct ip_moptions *imo, struct inpcb *inp)
 127 {
 128         struct ip *ip;
 129         struct ifnet *ifp = NULL;       /* keep compiler happy */
 130         struct mbuf *m;
 131         int hlen = sizeof(struct ip);
 132         int len, off, error = 0;
 133         struct sockaddr_in *dst = NULL; /* keep compiler happy */
 134         struct in_ifaddr *ia = NULL;
 135         int isbroadcast, sw_csum;
 136         struct in_addr pkt_dst;
 137         struct route iproute;
 138         struct m_tag *dn_mtag = NULL;
 139 #ifdef IPSEC
 140         struct secpolicy *sp = NULL;
 141         struct socket *so = inp ? inp->inp_socket : NULL;
 142 #endif
 143 #ifdef FAST_IPSEC
 144         struct m_tag *mtag;
 145         struct secpolicy *sp = NULL;
 146         struct tdb_ident *tdbi;
 147 #endif /* FAST_IPSEC */
 148         struct ip_fw_args args;
 149         int src_was_INADDR_ANY = 0;     /* as the name says... */
 150
 151         args.eh = NULL;
 152         args.rule = NULL;
 153         args.next_hop = NULL;
 154
 155         /* Grab info from MT_TAG mbufs prepended to the chain. */
 156         while (m0 != NULL && m0->m_type == MT_TAG) {
 157                 switch(m0->_m_tag_id) {
 158                 case PACKET_TAG_IPFORWARD:
 159                         args.next_hop = (struct sockaddr_in *)m0->m_data;
 160                         break;
 161                 default:
 162                         kprintf("ip_output: unrecognised MT_TAG tag %d\n",
 163                             m0->_m_tag_id);
 164                         break;
 165                 }
 166                 m0 = m0->m_next;
 167         }
 168         m = m0;
 169         M_ASSERTPKTHDR(m);
 170
 171         /* Extract info from dummynet tag */
 172         dn_mtag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL);
 173         if (dn_mtag != NULL) {
 174                 struct dn_pkt *dn_pkt = m_tag_data(dn_mtag);
 175
 176                 /*
 177                  * The packet was already tagged, so part of the
 178                  * processing was already done, and we need to go down.
 179                  * Get parameters from the tag.
 180                  */
 181                 args.rule = dn_pkt->dn_priv;
 182                 opt = NULL;
 183                 ro = &dn_pkt->ro;
 184                 imo = NULL;
 185                 dst = dn_pkt->dn_dst;
 186                 ifp = dn_pkt->ifp;
 187                 flags = dn_pkt->flags;
 188
 189                 /*
 190                  * Don't delete the dummynet tag here, just unlink it,
 191                  * since some local variables (like 'ro' and 'dst') are
 192                  * still referencing certain parts of it.
 193                  * The dummynet tag will be freed at the end of the
 194                  * output process.
 195                  */
 196                 m_tag_unlink(m, dn_mtag);
 197         }
 198
 199         if (ro == NULL) {
 200                 ro = &iproute;
 201                 bzero(ro, sizeof *ro);
 202         }
 203
 204         if (args.rule != NULL) {        /* dummynet already saw us */
 205                 ip = mtod(m, struct ip *);
 206                 hlen = IP_VHL_HL(ip->ip_vhl) << 2 ;
 207                 if (ro->ro_rt)
 208                         ia = ifatoia(ro->ro_rt->rt_ifa);
 209                 goto sendit;
 210         }
 211
 212         if (opt) {
 213                 len = 0;
 214                 m = ip_insertoptions(m, opt, &len);
 215                 if (len != 0)
 216                         hlen = len;
 217         }
 218         ip = mtod(m, struct ip *);
 219         pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst;
 220
 221         /*
 222          * Fill in IP header.
 223          */
 224         if (!(flags & (IP_FORWARDING|IP_RAWOUTPUT))) {
 225                 ip->ip_vhl = IP_MAKE_VHL(IPVERSION, hlen >> 2);
 226                 ip->ip_off &= IP_DF;
 227                 ip->ip_id = ip_newid();
 228                 ipstat.ips_localout++;
 229         } else {
 230                 hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 231         }
 232
 233         dst = (struct sockaddr_in *)&ro->ro_dst;
 234         /*
 235          * If there is a cached route,
 236          * check that it is to the same destination
 237          * and is still up.  If not, free it and try again.
 238          * The address family should also be checked in case of sharing the
 239          * cache with IPv6.
 240          */
 241         if (ro->ro_rt &&
 242             (!(ro->ro_rt->rt_flags & RTF_UP) ||
 243              dst->sin_family != AF_INET ||
 244              dst->sin_addr.s_addr != pkt_dst.s_addr)) {
 245                 rtfree(ro->ro_rt);
 246                 ro->ro_rt = (struct rtentry *)NULL;
 247         }
 248         if (ro->ro_rt == NULL) {
 249                 bzero(dst, sizeof *dst);
 250                 dst->sin_family = AF_INET;
 251                 dst->sin_len = sizeof *dst;
 252                 dst->sin_addr = pkt_dst;
 253         }
 254         /*
 255          * If routing to interface only,
 256          * short circuit routing lookup.
 257          */
 258         if (flags & IP_ROUTETOIF) {
 259                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL &&
 260                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) {
 261                         ipstat.ips_noroute++;
 262                         error = ENETUNREACH;
 263                         goto bad;
 264                 }
 265                 ifp = ia->ia_ifp;
 266                 ip->ip_ttl = 1;
 267                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 268         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 269                    imo != NULL && imo->imo_multicast_ifp != NULL) {
 270                 /*
 271                  * Bypass the normal routing lookup for multicast
 272                  * packets if the interface is specified.
 273                  */
 274                 ifp = imo->imo_multicast_ifp;
 275                 IFP_TO_IA(ifp, ia);
 276                 isbroadcast = 0;        /* fool gcc */
 277         } else {
 278                 /*
 279                  * If this is the case, we probably don't want to allocate
 280                  * a protocol-cloned route since we didn't get one from the
 281                  * ULP.  This lets TCP do its thing, while not burdening
 282                  * forwarding or ICMP with the overhead of cloning a route.
 283                  * Of course, we still want to do any cloning requested by
 284                  * the link layer, as this is probably required in all cases
 285                  * for correct operation (as it is for ARP).
 286                  */
 287                 if (ro->ro_rt == NULL)
 288                         rtalloc_ign(ro, RTF_PRCLONING);
 289                 if (ro->ro_rt == NULL) {
 290                         ipstat.ips_noroute++;
 291                         error = EHOSTUNREACH;
 292                         goto bad;
 293                 }
 294                 ia = ifatoia(ro->ro_rt->rt_ifa);
 295                 ifp = ro->ro_rt->rt_ifp;
 296                 ro->ro_rt->rt_use++;
 297                 if (ro->ro_rt->rt_flags & RTF_GATEWAY)
 298                         dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway;
 299                 if (ro->ro_rt->rt_flags & RTF_HOST)
 300                         isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST);
 301                 else
 302                         isbroadcast = in_broadcast(dst->sin_addr, ifp);
 303         }
 304         if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) {
 305                 struct in_multi *inm;
 306
 307                 m->m_flags |= M_MCAST;
 308                 /*
 309                  * IP destination address is multicast.  Make sure "dst"
 310                  * still points to the address in "ro".  (It may have been
 311                  * changed to point to a gateway address, above.)
 312                  */
 313                 dst = (struct sockaddr_in *)&ro->ro_dst;
 314                 /*
 315                  * See if the caller provided any multicast options
 316                  */
 317                 if (imo != NULL) {
 318                         ip->ip_ttl = imo->imo_multicast_ttl;
 319                         if (imo->imo_multicast_vif != -1)
 320                                 ip->ip_src.s_addr =
 321                                     ip_mcast_src ?
 322                                     ip_mcast_src(imo->imo_multicast_vif) :
 323                                     INADDR_ANY;
 324                 } else
 325                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 326                 /*
 327                  * Confirm that the outgoing interface supports multicast.
 328                  */
 329                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 330                         if (!(ifp->if_flags & IFF_MULTICAST)) {
 331                                 ipstat.ips_noroute++;
 332                                 error = ENETUNREACH;
 333                                 goto bad;
 334                         }
 335                 }
 336                 /*
 337                  * If source address not specified yet, use address
 338                  * of outgoing interface.
 339                  */
 340                 if (ip->ip_src.s_addr == INADDR_ANY) {
 341                         /* Interface may have no addresses. */
 342                         if (ia != NULL)
 343                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 344                 }
 345
 346                 IN_LOOKUP_MULTI(pkt_dst, ifp, inm);
 347                 if (inm != NULL &&
 348                    (imo == NULL || imo->imo_multicast_loop)) {
 349                         /*
 350                          * If we belong to the destination multicast group
 351                          * on the outgoing interface, and the caller did not
 352                          * forbid loopback, loop back a copy.
 353                          */
 354                         ip_mloopback(ifp, m, dst, hlen);
 355                 }
 356                 else {
 357                         /*
 358                          * If we are acting as a multicast router, perform
 359                          * multicast forwarding as if the packet had just
 360                          * arrived on the interface to which we are about
 361                          * to send.  The multicast forwarding function
 362                          * recursively calls this function, using the
 363                          * IP_FORWARDING flag to prevent infinite recursion.
 364                          *
 365                          * Multicasts that are looped back by ip_mloopback(),
 366                          * above, will be forwarded by the ip_input() routine,
 367                          * if necessary.
 368                          */
 369                         if (ip_mrouter && !(flags & IP_FORWARDING)) {
 370                                 /*
 371                                  * If rsvp daemon is not running, do not
 372                                  * set ip_moptions. This ensures that the packet
 373                                  * is multicast and not just sent down one link
 374                                  * as prescribed by rsvpd.
 375                                  */
 376                                 if (!rsvp_on)
 377                                         imo = NULL;
 378                                 if (ip_mforward &&
 379                                     ip_mforward(ip, ifp, m, imo) != 0) {
 380                                         m_freem(m);
 381                                         goto done;
 382                                 }
 383                         }
 384                 }
 385
 386                 /*
 387                  * Multicasts with a time-to-live of zero may be looped-
 388                  * back, above, but must not be transmitted on a network.
 389                  * Also, multicasts addressed to the loopback interface
 390                  * are not sent -- the above call to ip_mloopback() will
 391                  * loop back a copy if this host actually belongs to the
 392                  * destination group on the loopback interface.
 393                  */
 394                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 395                         m_freem(m);
 396                         goto done;
 397                 }
 398
 399                 goto sendit;
 400         }
 401 #ifndef notdef
 402         /*
 403          * If the source address is not specified yet, use the address
 404          * of the outoing interface. In case, keep note we did that, so
 405          * if the the firewall changes the next-hop causing the output
 406          * interface to change, we can fix that.
 407          */
 408         if (ip->ip_src.s_addr == INADDR_ANY) {
 409                 /* Interface may have no addresses. */
 410                 if (ia != NULL) {
 411                         ip->ip_src = IA_SIN(ia)->sin_addr;
 412                         src_was_INADDR_ANY = 1;
 413                 }
 414         }
 415 #endif /* notdef */
 416 #ifdef ALTQ
 417         /*
 418          * Disable packet drop hack.
 419          * Packetdrop should be done by queueing.
 420          */
 421 #else /* !ALTQ */
 422         /*
 423          * Verify that we have any chance at all of being able to queue
 424          *      the packet or packet fragments
 425          */
 426         if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >=
 427                 ifp->if_snd.ifq_maxlen) {
 428                         error = ENOBUFS;
 429                         ipstat.ips_odropped++;
 430                         goto bad;
 431         }
 432 #endif /* !ALTQ */
 433
 434         /*
 435          * Look for broadcast address and
 436          * verify user is allowed to send
 437          * such a packet.
 438          */
 439         if (isbroadcast) {
 440                 if (!(ifp->if_flags & IFF_BROADCAST)) {
 441                         error = EADDRNOTAVAIL;
 442                         goto bad;
 443                 }
 444                 if (!(flags & IP_ALLOWBROADCAST)) {
 445                         error = EACCES;
 446                         goto bad;
 447                 }
 448                 /* don't allow broadcast messages to be fragmented */
 449                 if (ip->ip_len > ifp->if_mtu) {
 450                         error = EMSGSIZE;
 451                         goto bad;
 452                 }
 453                 m->m_flags |= M_BCAST;
 454         } else {
 455                 m->m_flags &= ~M_BCAST;
 456         }
 457
 458 sendit:
 459 #ifdef IPSEC
 460         /* get SP for this packet */
 461         if (so == NULL)
 462                 sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, flags, &error);
 463         else
 464                 sp = ipsec4_getpolicybysock(m, IPSEC_DIR_OUTBOUND, so, &error);
 465
 466         if (sp == NULL) {
 467                 ipsecstat.out_inval++;
 468                 goto bad;
 469         }
 470
 471         error = 0;
 472
 473         /* check policy */
 474         switch (sp->policy) {
 475         case IPSEC_POLICY_DISCARD:
 476                 /*
 477                  * This packet is just discarded.
 478                  */
 479                 ipsecstat.out_polvio++;
 480                 goto bad;
 481
 482         case IPSEC_POLICY_BYPASS:
 483         case IPSEC_POLICY_NONE:
 484                 /* no need to do IPsec. */
 485                 goto skip_ipsec;
 486
 487         case IPSEC_POLICY_IPSEC:
 488                 if (sp->req == NULL) {
 489                         /* acquire a policy */
 490                         error = key_spdacquire(sp);
 491                         goto bad;
 492                 }
 493                 break;
 494
 495         case IPSEC_POLICY_ENTRUST:
 496         default:
 497                 kprintf("ip_output: Invalid policy found. %d\n", sp->policy);
 498         }
 499     {
 500         struct ipsec_output_state state;
 501         bzero(&state, sizeof state);
 502         state.m = m;
 503         if (flags & IP_ROUTETOIF) {
 504                 state.ro = &iproute;
 505                 bzero(&iproute, sizeof iproute);
 506         } else
 507                 state.ro = ro;
 508         state.dst = (struct sockaddr *)dst;
 509
 510         ip->ip_sum = 0;
 511
 512         /*
 513          * XXX
 514          * delayed checksums are not currently compatible with IPsec
 515          */
 516         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 517                 in_delayed_cksum(m);
 518                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 519         }
 520
 521         ip->ip_len = htons(ip->ip_len);
 522         ip->ip_off = htons(ip->ip_off);
 523
 524         error = ipsec4_output(&state, sp, flags);
 525
 526         m = state.m;
 527         if (flags & IP_ROUTETOIF) {
 528                 /*
 529                  * if we have tunnel mode SA, we may need to ignore
 530                  * IP_ROUTETOIF.
 531                  */
 532                 if (state.ro != &iproute || state.ro->ro_rt != NULL) {
 533                         flags &= ~IP_ROUTETOIF;
 534                         ro = state.ro;
 535                 }
 536         } else
 537                 ro = state.ro;
 538         dst = (struct sockaddr_in *)state.dst;
 539         if (error) {
 540                 /* mbuf is already reclaimed in ipsec4_output. */
 541                 m0 = NULL;
 542                 switch (error) {
 543                 case EHOSTUNREACH:
 544                 case ENETUNREACH:
 545                 case EMSGSIZE:
 546                 case ENOBUFS:
 547                 case ENOMEM:
 548                         break;
 549                 default:
 550                         kprintf("ip4_output (ipsec): error code %d\n", error);
 551                         /*fall through*/
 552                 case ENOENT:
 553                         /* don't show these error codes to the user */
 554                         error = 0;
 555                         break;
 556                 }
 557                 goto bad;
 558         }
 559     }
 560
 561         /* be sure to update variables that are affected by ipsec4_output() */
 562         ip = mtod(m, struct ip *);
 563 #ifdef _IP_VHL
 564         hlen = IP_VHL_HL(ip->ip_vhl) << 2;
 565 #else
 566         hlen = ip->ip_hl << 2;
 567 #endif
 568         if (ro->ro_rt == NULL) {
 569                 if (!(flags & IP_ROUTETOIF)) {
 570                         kprintf("ip_output: "
 571                                 "can't update route after IPsec processing\n");
 572                         error = EHOSTUNREACH;   /*XXX*/
 573                         goto bad;
 574                 }
 575         } else {
 576                 ia = ifatoia(ro->ro_rt->rt_ifa);
 577                 ifp = ro->ro_rt->rt_ifp;
 578         }
 579
 580         /* make it flipped, again. */
 581         ip->ip_len = ntohs(ip->ip_len);
 582         ip->ip_off = ntohs(ip->ip_off);
 583 skip_ipsec:
 584 #endif /*IPSEC*/
 585 #ifdef FAST_IPSEC
 586         /*
 587          * Check the security policy (SP) for the packet and, if
 588          * required, do IPsec-related processing.  There are two
 589          * cases here; the first time a packet is sent through
 590          * it will be untagged and handled by ipsec4_checkpolicy.
 591          * If the packet is resubmitted to ip_output (e.g. after
 592          * AH, ESP, etc. processing), there will be a tag to bypass
 593          * the lookup and related policy checking.
 594          */
 595         mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL);
 596         crit_enter();
 597         if (mtag != NULL) {
 598                 tdbi = (struct tdb_ident *)m_tag_data(mtag);
 599                 sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND);
 600                 if (sp == NULL)
 601                         error = -EINVAL;        /* force silent drop */
 602                 m_tag_delete(m, mtag);
 603         } else {
 604                 sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags,
 605                                         &error, inp);
 606         }
 607         /*
 608          * There are four return cases:
 609          *    sp != NULL                    apply IPsec policy
 610          *    sp == NULL, error == 0        no IPsec handling needed
 611          *    sp == NULL, error == -EINVAL  discard packet w/o error
 612          *    sp == NULL, error != 0        discard packet, report error
 613          */
 614         if (sp != NULL) {
 615                 /* Loop detection, check if ipsec processing already done */
 616                 KASSERT(sp->req != NULL, ("ip_output: no ipsec request"));
 617                 for (mtag = m_tag_first(m); mtag != NULL;
 618                      mtag = m_tag_next(m, mtag)) {
 619                         if (mtag->m_tag_cookie != MTAG_ABI_COMPAT)
 620                                 continue;
 621                         if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE &&
 622                             mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED)
 623                                 continue;
 624                         /*
 625                          * Check if policy has an SA associated with it.
 626                          * This can happen when an SP has yet to acquire
 627                          * an SA; e.g. on first reference.  If it occurs,
 628                          * then we let ipsec4_process_packet do its thing.
 629                          */
 630                         if (sp->req->sav == NULL)
 631                                 break;
 632                         tdbi = (struct tdb_ident *)m_tag_data(mtag);
 633                         if (tdbi->spi == sp->req->sav->spi &&
 634                             tdbi->proto == sp->req->sav->sah->saidx.proto &&
 635                             bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst,
 636                                  sizeof(union sockaddr_union)) == 0) {
 637                                 /*
 638                                  * No IPsec processing is needed, free
 639                                  * reference to SP.
 640                                  *
 641                                  * NB: null pointer to avoid free at
 642                                  *     done: below.
 643                                  */
 644                                 KEY_FREESP(&sp), sp = NULL;
 645                                 crit_exit();
 646                                 goto spd_done;
 647                         }
 648                 }
 649
 650                 /*
 651                  * Do delayed checksums now because we send before
 652                  * this is done in the normal processing path.
 653                  */
 654                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 655                         in_delayed_cksum(m);
 656                         m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 657                 }
 658
 659                 ip->ip_len = htons(ip->ip_len);
 660                 ip->ip_off = htons(ip->ip_off);
 661
 662                 /* NB: callee frees mbuf */
 663                 error = ipsec4_process_packet(m, sp->req, flags, 0);
 664                 /*
 665                  * Preserve KAME behaviour: ENOENT can be returned
 666                  * when an SA acquire is in progress.  Don't propagate
 667                  * this to user-level; it confuses applications.
 668                  *
 669                  * XXX this will go away when the SADB is redone.
 670                  */
 671                 if (error == ENOENT)
 672                         error = 0;
 673                 crit_exit();
 674                 goto done;
 675         } else {
 676                 crit_exit();
 677
 678                 if (error != 0) {
 679                         /*
 680                          * Hack: -EINVAL is used to signal that a packet
 681                          * should be silently discarded.  This is typically
 682                          * because we asked key management for an SA and
 683                          * it was delayed (e.g. kicked up to IKE).
 684                          */
 685                         if (error == -EINVAL)
 686                                 error = 0;
 687                         goto bad;
 688                 } else {
 689                         /* No IPsec processing for this packet. */
 690                 }
 691 #ifdef notyet
 692                 /*
 693                  * If deferred crypto processing is needed, check that
 694                  * the interface supports it.
 695                  */
 696                 mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL);
 697                 if (mtag != NULL && !(ifp->if_capenable & IFCAP_IPSEC)) {
 698                         /* notify IPsec to do its own crypto */
 699                         ipsp_skipcrypto_unmark((struct tdb_ident *)m_tag_data(mtag));
 700                         error = EHOSTUNREACH;
 701                         goto bad;
 702                 }
 703 #endif
 704         }
 705 spd_done:
 706 #endif /* FAST_IPSEC */
 707         /*
 708          * IpHack's section.
 709          * - Xlate: translate packet's addr/port (NAT).
 710          * - Firewall: deny/allow/etc.
 711          * - Wrap: fake packet's addr/port <unimpl.>
 712          * - Encapsulate: put it in another IP and send out. <unimp.>
 713          */
 714
 715         /*
 716          * Run through list of hooks for output packets.
 717          */
 718         if (pfil_has_hooks(&inet_pfil_hook)) {
 719                 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT);
 720                 if (error != 0 || m == NULL)
 721                         goto done;
 722                 ip = mtod(m, struct ip *);
 723         }
 724
 725         /*
 726          * Check with the firewall...
 727          * but not if we are already being fwd'd from a firewall.
 728          */
 729         if (fw_enable && IPFW_LOADED && !args.next_hop) {
 730                 struct sockaddr_in *old = dst;
 731
 732                 args.m = m;
 733                 args.next_hop = dst;
 734                 args.oif = ifp;
 735                 off = ip_fw_chk_ptr(&args);
 736                 m = args.m;
 737                 dst = args.next_hop;
 738
 739                 /*
 740                  * On return we must do the following:
 741                  * m == NULL    -> drop the pkt (old interface, deprecated)
 742                  * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface)
 743                  * 1<=off<= 0xffff              -> DIVERT
 744                  * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe
 745                  * (off & IP_FW_PORT_TEE_FLAG)  -> TEE the packet
 746                  * dst != old                   -> IPFIREWALL_FORWARD
 747                  * off==0, dst==old             -> accept
 748                  * If some of the above modules are not compiled in, then
 749                  * we should't have to check the corresponding condition
 750                  * (because the ipfw control socket should not accept
 751                  * unsupported rules), but better play safe and drop
 752                  * packets in case of doubt.
 753                  */
 754                 if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) {
 755                         if (m)
 756                                 m_freem(m);
 757                         error = EACCES;
 758                         goto done;
 759                 }
 760                 ip = mtod(m, struct ip *);
 761                 if (off == 0 && dst == old)             /* common case */
 762                         goto pass;
 763                 if (off & IP_FW_PORT_DYNT_FLAG) {
 764                         /*
 765                          * pass the pkt to dummynet. Need to include
 766                          * pipe number, m, ifp, ro, dst because these are
 767                          * not recomputed in the next pass.
 768                          * All other parameters have been already used and
 769                          * so they are not needed anymore.
 770                          * XXX note: if the ifp or ro entry are deleted
 771                          * while a pkt is in dummynet, we are in trouble!
 772                          */
 773                         args.ro = ro;
 774                         args.dst = dst;
 775                         args.flags = flags;
 776
 777                         error = 0;
 778                         ip_fw_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT, &args);
 779                         goto done;
 780                 }
 781 #ifdef IPDIVERT
 782                 if (off != 0 && !(off & IP_FW_PORT_DYNT_FLAG)) {
 783                         struct mbuf *clone = NULL;
 784
 785                         /* Clone packet if we're doing a 'tee' */
 786                         if ((off & IP_FW_PORT_TEE_FLAG))
 787                                 clone = m_dup(m, MB_DONTWAIT);
 788
 789                         /*
 790                          * XXX
 791                          * delayed checksums are not currently compatible
 792                          * with divert sockets.
 793                          */
 794                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 795                                 in_delayed_cksum(m);
 796                                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 797                         }
 798
 799                         /* Restore packet header fields to original values */
 800                         ip->ip_len = htons(ip->ip_len);
 801                         ip->ip_off = htons(ip->ip_off);
 802
 803                         /* Deliver packet to divert input routine */
 804                         divert_packet(m, 0, off & 0xffff);
 805
 806                         /* If 'tee', continue with original packet */
 807                         if (clone != NULL) {
 808                                 m = clone;
 809                                 ip = mtod(m, struct ip *);
 810                                 goto pass;
 811                         }
 812                         goto done;
 813                 }
 814 #endif
 815
 816                 /* IPFIREWALL_FORWARD */
 817                 /*
 818                  * Check dst to make sure it is directly reachable on the
 819                  * interface we previously thought it was.
 820                  * If it isn't (which may be likely in some situations) we have
 821                  * to re-route it (ie, find a route for the next-hop and the
 822                  * associated interface) and set them here. This is nested
 823                  * forwarding which in most cases is undesirable, except where
 824                  * such control is nigh impossible. So we do it here.
 825                  * And I'm babbling.
 826                  */
 827                 if (off == 0 && old != dst) { /* FORWARD, dst has changed */
 828 #if 0
 829                         /*
 830                          * XXX To improve readability, this block should be
 831                          * changed into a function call as below:
 832                          */
 833                         error = ip_ipforward(&m, &dst, &ifp);
 834                         if (error)
 835                                 goto bad;
 836                         if (m == NULL) /* ip_input consumed the mbuf */
 837                                 goto done;
 838 #else
 839                         struct in_ifaddr *ia;
 840
 841                         /*
 842                          * XXX sro_fwd below is static, and a pointer
 843                          * to it gets passed to routines downstream.
 844                          * This could have surprisingly bad results in
 845                          * practice, because its content is overwritten
 846                          * by subsequent packets.
 847                          */
 848                         /* There must be a better way to do this next line... */
 849                         static struct route sro_fwd;
 850                         struct route *ro_fwd = &sro_fwd;
 851
 852 #if 0
 853                         print_ip("IPFIREWALL_FORWARD: New dst ip: ",
 854                             dst->sin_addr, "\n");
 855 #endif
 856
 857                         /*
 858                          * We need to figure out if we have been forwarded
 859                          * to a local socket. If so, then we should somehow
 860                          * "loop back" to ip_input, and get directed to the
 861                          * PCB as if we had received this packet. This is
 862                          * because it may be dificult to identify the packets
 863                          * you want to forward until they are being output
 864                          * and have selected an interface. (e.g. locally
 865                          * initiated packets) If we used the loopback inteface,
 866                          * we would not be able to control what happens
 867                          * as the packet runs through ip_input() as
 868                          * it is done through a ISR.
 869                          */
 870                         LIST_FOREACH(ia, INADDR_HASH(dst->sin_addr.s_addr),
 871                                      ia_hash) {
 872                                 /*
 873                                  * If the addr to forward to is one
 874                                  * of ours, we pretend to
 875                                  * be the destination for this packet.
 876                                  */
 877                                 if (IA_SIN(ia)->sin_addr.s_addr ==
 878                                                  dst->sin_addr.s_addr)
 879                                         break;
 880                         }
 881                         if (ia != NULL) {    /* tell ip_input "dont filter" */
 882                                 struct m_hdr tag;
 883
 884                                 tag.mh_type = MT_TAG;
 885                                 tag.mh_flags = PACKET_TAG_IPFORWARD;
 886                                 tag.mh_data = (caddr_t)args.next_hop;
 887                                 tag.mh_next = m;
 888
 889                                 if (m->m_pkthdr.rcvif == NULL)
 890                                         m->m_pkthdr.rcvif = ifunit("lo0");
 891                                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 892                                         m->m_pkthdr.csum_flags |=
 893                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 894                                         m->m_pkthdr.csum_data = 0xffff;
 895                                 }
 896                                 m->m_pkthdr.csum_flags |=
 897                                     CSUM_IP_CHECKED | CSUM_IP_VALID;
 898                                 ip->ip_len = htons(ip->ip_len);
 899                                 ip->ip_off = htons(ip->ip_off);
 900                                 ip_input((struct mbuf *)&tag);
 901                                 goto done;
 902                         }
 903                         /* Some of the logic for this was nicked from above.
 904                          *
 905                          * This rewrites the cached route in a local PCB.
 906                          * Is this what we want to do?
 907                          */
 908                         bcopy(dst, &ro_fwd->ro_dst, sizeof *dst);
 909                         ro_fwd->ro_rt = NULL;
 910
 911                         rtalloc_ign(ro_fwd, RTF_PRCLONING);
 912                         if (ro_fwd->ro_rt == NULL) {
 913                                 ipstat.ips_noroute++;
 914                                 error = EHOSTUNREACH;
 915                                 goto bad;
 916                         }
 917
 918                         ia = ifatoia(ro_fwd->ro_rt->rt_ifa);
 919                         ifp = ro_fwd->ro_rt->rt_ifp;
 920                         ro_fwd->ro_rt->rt_use++;
 921                         if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY)
 922                                 dst = (struct sockaddr_in *)
 923                                     ro_fwd->ro_rt->rt_gateway;
 924                         if (ro_fwd->ro_rt->rt_flags & RTF_HOST)
 925                                 isbroadcast =
 926                                     (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST);
 927                         else
 928                                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 929                         if (ro->ro_rt != NULL)
 930                                 rtfree(ro->ro_rt);
 931                         ro->ro_rt = ro_fwd->ro_rt;
 932                         dst = (struct sockaddr_in *)&ro_fwd->ro_dst;
 933
 934 #endif  /* ... block to be put into a function */
 935                         /*
 936                          * If we added a default src ip earlier,
 937                          * which would have been gotten from the-then
 938                          * interface, do it again, from the new one.
 939                          */
 940                         if (src_was_INADDR_ANY)
 941                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 942                         goto pass ;
 943                 }
 944
 945                 /*
 946                  * if we get here, none of the above matches, and
 947                  * we have to drop the pkt
 948                  */
 949                 m_freem(m);
 950                 error = EACCES; /* not sure this is the right error msg */
 951                 goto done;
 952         }
 953
 954 pass:
 955         /* 127/8 must not appear on wire - RFC1122. */
 956         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 957             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 958                 if (!(ifp->if_flags & IFF_LOOPBACK)) {
 959                         ipstat.ips_badaddr++;
 960                         error = EADDRNOTAVAIL;
 961                         goto bad;
 962                 }
 963         }
 964
 965         m->m_pkthdr.csum_flags |= CSUM_IP;
 966         sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist;
 967         if (sw_csum & CSUM_DELAY_DATA) {
 968                 in_delayed_cksum(m);
 969                 sw_csum &= ~CSUM_DELAY_DATA;
 970         }
 971         m->m_pkthdr.csum_flags &= ifp->if_hwassist;
 972
 973         /*
 974          * If small enough for interface, or the interface will take
 975          * care of the fragmentation for us, can just send directly.
 976          */
 977         if (ip->ip_len <= ifp->if_mtu || ((ifp->if_hwassist & CSUM_FRAGMENT) &&
 978             !(ip->ip_off & IP_DF))) {
 979                 ip->ip_len = htons(ip->ip_len);
 980                 ip->ip_off = htons(ip->ip_off);
 981                 ip->ip_sum = 0;
 982                 if (sw_csum & CSUM_DELAY_IP) {
 983                         if (ip->ip_vhl == IP_VHL_BORING) {
 984                                 ip->ip_sum = in_cksum_hdr(ip);
 985                         } else {
 986                                 ip->ip_sum = in_cksum(m, hlen);
 987                         }
 988                 }
 989
 990                 /* Record statistics for this interface address. */
 991                 if (!(flags & IP_FORWARDING) && ia) {
 992                         ia->ia_ifa.if_opackets++;
 993                         ia->ia_ifa.if_obytes += m->m_pkthdr.len;
 994                 }
 995
 996 #ifdef IPSEC
 997                 /* clean ipsec history once it goes out of the node */
 998                 ipsec_delaux(m);
 999 #endif
1000
1001 #ifdef MBUF_STRESS_TEST
1002                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) {
1003                         struct mbuf *m1, *m2;
1004                         int length, tmp;
1005
1006                         tmp = length = m->m_pkthdr.len;
1007
1008                         while ((length -= mbuf_frag_size) >= 1) {
1009                                 m1 = m_split(m, length, MB_DONTWAIT);
1010                                 if (m1 == NULL)
1011                                         break;
1012                                 m2 = m;
1013                                 while (m2->m_next != NULL)
1014                                         m2 = m2->m_next;
1015                                 m2->m_next = m1;
1016                         }
1017                         m->m_pkthdr.len = tmp;
1018                 }
1019 #endif
1020                 lwkt_serialize_enter(ifp->if_serializer);
1021                 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst,
1022                                           ro->ro_rt);
1023                 lwkt_serialize_exit(ifp->if_serializer);
1024                 goto done;
1025         }
1026
1027         if (ip->ip_off & IP_DF) {
1028                 error = EMSGSIZE;
1029                 /*
1030                  * This case can happen if the user changed the MTU
1031                  * of an interface after enabling IP on it.  Because
1032                  * most netifs don't keep track of routes pointing to
1033                  * them, there is no way for one to update all its
1034                  * routes when the MTU is changed.
1035                  */
1036                 if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) &&
1037                     !(ro->ro_rt->rt_rmx.rmx_locks & RTV_MTU) &&
1038                     (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) {
1039                         ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu;
1040                 }
1041                 ipstat.ips_cantfrag++;
1042                 goto bad;
1043         }
1044
1045         /*
1046          * Too large for interface; fragment if possible. If successful,
1047          * on return, m will point to a list of packets to be sent.
1048          */
1049         error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum);
1050         if (error)
1051                 goto bad;
1052         for (; m; m = m0) {
1053                 m0 = m->m_nextpkt;
1054                 m->m_nextpkt = NULL;
1055 #ifdef IPSEC
1056                 /* clean ipsec history once it goes out of the node */
1057                 ipsec_delaux(m);
1058 #endif
1059                 if (error == 0) {
1060                         /* Record statistics for this interface address. */
1061                         if (ia != NULL) {
1062                                 ia->ia_ifa.if_opackets++;
1063                                 ia->ia_ifa.if_obytes += m->m_pkthdr.len;
1064                         }
1065                         lwkt_serialize_enter(ifp->if_serializer);
1066                         error = (*ifp->if_output)(ifp, m,
1067                                                   (struct sockaddr *)dst,
1068                                                   ro->ro_rt);
1069                         lwkt_serialize_exit(ifp->if_serializer);
1070                 } else {
1071                         m_freem(m);
1072                 }
1073         }
1074
1075         if (error == 0)
1076                 ipstat.ips_fragmented++;
1077
1078 done:
1079         if (ro == &iproute && ro->ro_rt != NULL) {
1080                 RTFREE(ro->ro_rt);
1081                 ro->ro_rt = NULL;
1082         }
1083 #ifdef IPSEC
1084         if (sp != NULL) {
1085                 KEYDEBUG(KEYDEBUG_IPSEC_STAMP,
1086                         kprintf("DP ip_output call free SP:%p\n", sp));
1087                 key_freesp(sp);
1088         }
1089 #endif
1090 #ifdef FAST_IPSEC
1091         if (sp != NULL)
1092                 KEY_FREESP(&sp);
1093 #endif
1094         if (dn_mtag != NULL)
1095                 m_tag_free(dn_mtag);
1096
1097         return (error);
1098 bad:
1099         m_freem(m);
1100         goto done;
1101 }
1102
1103 /*
1104  * Create a chain of fragments which fit the given mtu. m_frag points to the
1105  * mbuf to be fragmented; on return it points to the chain with the fragments.
1106  * Return 0 if no error. If error, m_frag may contain a partially built
1107  * chain of fragments that should be freed by the caller.
1108  *
1109  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
1110  * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP).
1111  */
1112 int
1113 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
1114             u_long if_hwassist_flags, int sw_csum)
1115 {
1116         int error = 0;
1117         int hlen = IP_VHL_HL(ip->ip_vhl) << 2;
1118         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
1119         int off;
1120         struct mbuf *m0 = *m_frag;      /* the original packet          */
1121         int firstlen;
1122         struct mbuf **mnext;
1123         int nfrags;
1124
1125         if (ip->ip_off & IP_DF) {       /* Fragmentation not allowed */
1126                 ipstat.ips_cantfrag++;
1127                 return EMSGSIZE;
1128         }
1129
1130         /*
1131          * Must be able to put at least 8 bytes per fragment.
1132          */
1133         if (len < 8)
1134                 return EMSGSIZE;
1135
1136         /*
1137          * If the interface will not calculate checksums on
1138          * fragmented packets, then do it here.
1139          */
1140         if ((m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) &&
1141             !(if_hwassist_flags & CSUM_IP_FRAGS)) {
1142                 in_delayed_cksum(m0);
1143                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1144         }
1145
1146         if (len > PAGE_SIZE) {
1147                 /*
1148                  * Fragment large datagrams such that each segment
1149                  * contains a multiple of PAGE_SIZE amount of data,
1150                  * plus headers. This enables a receiver to perform
1151                  * page-flipping zero-copy optimizations.
1152                  *
1153                  * XXX When does this help given that sender and receiver
1154                  * could have different page sizes, and also mtu could
1155                  * be less than the receiver's page size ?
1156                  */
1157                 int newlen;
1158                 struct mbuf *m;
1159
1160                 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next)
1161                         off += m->m_len;
1162
1163                 /*
1164                  * firstlen (off - hlen) must be aligned on an
1165                  * 8-byte boundary
1166                  */
1167                 if (off < hlen)
1168                         goto smart_frag_failure;
1169                 off = ((off - hlen) & ~7) + hlen;
1170                 newlen = (~PAGE_MASK) & mtu;
1171                 if ((newlen + sizeof(struct ip)) > mtu) {
1172                         /* we failed, go back the default */
1173 smart_frag_failure:
1174                         newlen = len;
1175                         off = hlen + len;
1176                 }
1177                 len = newlen;
1178
1179         } else {
1180                 off = hlen + len;
1181         }
1182
1183         firstlen = off - hlen;
1184         mnext = &m0->m_nextpkt;         /* pointer to next packet */
1185
1186         /*
1187          * Loop through length of segment after first fragment,
1188          * make new header and copy data of each part and link onto chain.
1189          * Here, m0 is the original packet, m is the fragment being created.
1190          * The fragments are linked off the m_nextpkt of the original
1191          * packet, which after processing serves as the first fragment.
1192          */
1193         for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) {
1194                 struct ip *mhip;        /* ip header on the fragment */
1195                 struct mbuf *m;
1196                 int mhlen = sizeof(struct ip);
1197
1198                 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1199                 if (m == NULL) {
1200                         error = ENOBUFS;
1201                         ipstat.ips_odropped++;
1202                         goto done;
1203                 }
1204                 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG;
1205                 /*
1206                  * In the first mbuf, leave room for the link header, then
1207                  * copy the original IP header including options. The payload
1208                  * goes into an additional mbuf chain returned by m_copy().
1209                  */
1210                 m->m_data += max_linkhdr;
1211                 mhip = mtod(m, struct ip *);
1212                 *mhip = *ip;
1213                 if (hlen > sizeof(struct ip)) {
1214                         mhlen = ip_optcopy(ip, mhip) + sizeof(struct ip);
1215                         mhip->ip_vhl = IP_MAKE_VHL(IPVERSION, mhlen >> 2);
1216                 }
1217                 m->m_len = mhlen;
1218                 /* XXX do we need to add ip->ip_off below ? */
1219                 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off;
1220                 if (off + len >= ip->ip_len) {  /* last fragment */
1221                         len = ip->ip_len - off;
1222                         m->m_flags |= M_LASTFRAG;
1223                 } else
1224                         mhip->ip_off |= IP_MF;
1225                 mhip->ip_len = htons((u_short)(len + mhlen));
1226                 m->m_next = m_copy(m0, off, len);
1227                 if (m->m_next == NULL) {                /* copy failed */
1228                         m_free(m);
1229                         error = ENOBUFS;        /* ??? */
1230                         ipstat.ips_odropped++;
1231                         goto done;
1232                 }
1233                 m->m_pkthdr.len = mhlen + len;
1234                 m->m_pkthdr.rcvif = (struct ifnet *)NULL;
1235                 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags;
1236                 mhip->ip_off = htons(mhip->ip_off);
1237                 mhip->ip_sum = 0;
1238                 if (sw_csum & CSUM_DELAY_IP)
1239                         mhip->ip_sum = in_cksum(m, mhlen);
1240                 *mnext = m;
1241                 mnext = &m->m_nextpkt;
1242         }
1243         ipstat.ips_ofragments += nfrags;
1244
1245         /* set first marker for fragment chain */
1246         m0->m_flags |= M_FIRSTFRAG | M_FRAG;
1247         m0->m_pkthdr.csum_data = nfrags;
1248
1249         /*
1250          * Update first fragment by trimming what's been copied out
1251          * and updating header.
1252          */
1253         m_adj(m0, hlen + firstlen - ip->ip_len);
1254         m0->m_pkthdr.len = hlen + firstlen;
1255         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1256         ip->ip_off |= IP_MF;
1257         ip->ip_off = htons(ip->ip_off);
1258         ip->ip_sum = 0;
1259         if (sw_csum & CSUM_DELAY_IP)
1260                 ip->ip_sum = in_cksum(m0, hlen);
1261
1262 done:
1263         *m_frag = m0;
1264         return error;
1265 }
1266
1267 void
1268 in_delayed_cksum(struct mbuf *m)
1269 {
1270         struct ip *ip;
1271         u_short csum, offset;
1272
1273         ip = mtod(m, struct ip *);
1274         offset = IP_VHL_HL(ip->ip_vhl) << 2 ;
1275         csum = in_cksum_skip(m, ip->ip_len, offset);
1276         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
1277                 csum = 0xffff;
1278         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1279
1280         if (offset + sizeof(u_short) > m->m_len) {
1281                 kprintf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
1282                     m->m_len, offset, ip->ip_p);
1283                 /*
1284                  * XXX
1285                  * this shouldn't happen, but if it does, the
1286                  * correct behavior may be to insert the checksum
1287                  * in the existing chain instead of rearranging it.
1288                  */
1289                 m = m_pullup(m, offset + sizeof(u_short));
1290         }
1291         *(u_short *)(m->m_data + offset) = csum;
1292 }
1293
1294 /*
1295  * Insert IP options into preformed packet.
1296  * Adjust IP destination as required for IP source routing,
1297  * as indicated by a non-zero in_addr at the start of the options.
1298  *
1299  * XXX This routine assumes that the packet has no options in place.
1300  */
1301 static struct mbuf *
1302 ip_insertoptions(struct mbuf *m, struct mbuf *opt, int *phlen)
1303 {
1304         struct ipoption *p = mtod(opt, struct ipoption *);
1305         struct mbuf *n;
1306         struct ip *ip = mtod(m, struct ip *);
1307         unsigned optlen;
1308
1309         optlen = opt->m_len - sizeof p->ipopt_dst;
1310         if (optlen + (u_short)ip->ip_len > IP_MAXPACKET) {
1311                 *phlen = 0;
1312                 return (m);             /* XXX should fail */
1313         }
1314         if (p->ipopt_dst.s_addr)
1315                 ip->ip_dst = p->ipopt_dst;
1316         if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) {
1317                 MGETHDR(n, MB_DONTWAIT, MT_HEADER);
1318                 if (n == NULL) {
1319                         *phlen = 0;
1320                         return (m);
1321                 }
1322                 n->m_pkthdr.rcvif = (struct ifnet *)NULL;
1323                 n->m_pkthdr.len = m->m_pkthdr.len + optlen;
1324                 m->m_len -= sizeof(struct ip);
1325                 m->m_data += sizeof(struct ip);
1326                 n->m_next = m;
1327                 m = n;
1328                 m->m_len = optlen + sizeof(struct ip);
1329                 m->m_data += max_linkhdr;
1330                 memcpy(mtod(m, void *), ip, sizeof(struct ip));
1331         } else {
1332                 m->m_data -= optlen;
1333                 m->m_len += optlen;
1334                 m->m_pkthdr.len += optlen;
1335                 ovbcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
1336         }
1337         ip = mtod(m, struct ip *);
1338         bcopy(p->ipopt_list, ip + 1, optlen);
1339         *phlen = sizeof(struct ip) + optlen;
1340         ip->ip_vhl = IP_MAKE_VHL(IPVERSION, *phlen >> 2);
1341         ip->ip_len += optlen;
1342         return (m);
1343 }
1344
1345 /*
1346  * Copy options from ip to jp,
1347  * omitting those not copied during fragmentation.
1348  */
1349 int
1350 ip_optcopy(struct ip *ip, struct ip *jp)
1351 {
1352         u_char *cp, *dp;
1353         int opt, optlen, cnt;
1354
1355         cp = (u_char *)(ip + 1);
1356         dp = (u_char *)(jp + 1);
1357         cnt = (IP_VHL_HL(ip->ip_vhl) << 2) - sizeof(struct ip);
1358         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1359                 opt = cp[0];
1360                 if (opt == IPOPT_EOL)
1361                         break;
1362                 if (opt == IPOPT_NOP) {
1363                         /* Preserve for IP mcast tunnel's LSRR alignment. */
1364                         *dp++ = IPOPT_NOP;
1365                         optlen = 1;
1366                         continue;
1367                 }
1368
1369                 KASSERT(cnt >= IPOPT_OLEN + sizeof *cp,
1370                     ("ip_optcopy: malformed ipv4 option"));
1371                 optlen = cp[IPOPT_OLEN];
1372                 KASSERT(optlen >= IPOPT_OLEN + sizeof *cp && optlen <= cnt,
1373                     ("ip_optcopy: malformed ipv4 option"));
1374
1375                 /* bogus lengths should have been caught by ip_dooptions */
1376                 if (optlen > cnt)
1377                         optlen = cnt;
1378                 if (IPOPT_COPIED(opt)) {
1379                         bcopy(cp, dp, optlen);
1380                         dp += optlen;
1381                 }
1382         }
1383         for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++)
1384                 *dp++ = IPOPT_EOL;
1385         return (optlen);
1386 }
1387
1388 /*
1389  * IP socket option processing.
1390  */
1391 int
1392 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1393 {
1394         struct  inpcb *inp = so->so_pcb;
1395         int     error, optval;
1396
1397         error = optval = 0;
1398         if (sopt->sopt_level != IPPROTO_IP) {
1399                 return (EINVAL);
1400         }
1401
1402         switch (sopt->sopt_dir) {
1403         case SOPT_SET:
1404                 switch (sopt->sopt_name) {
1405                 case IP_OPTIONS:
1406 #ifdef notyet
1407                 case IP_RETOPTS:
1408 #endif
1409                 {
1410                         struct mbuf *m;
1411                         if (sopt->sopt_valsize > MLEN) {
1412                                 error = EMSGSIZE;
1413                                 break;
1414                         }
1415                         MGET(m, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_HEADER);
1416                         if (m == NULL) {
1417                                 error = ENOBUFS;
1418                                 break;
1419                         }
1420                         m->m_len = sopt->sopt_valsize;
1421                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1422                                             m->m_len);
1423
1424                         return (ip_pcbopts(sopt->sopt_name, &inp->inp_options,
1425                                            m));
1426                 }
1427
1428                 case IP_TOS:
1429                 case IP_TTL:
1430                 case IP_MINTTL:
1431                 case IP_RECVOPTS:
1432                 case IP_RECVRETOPTS:
1433                 case IP_RECVDSTADDR:
1434                 case IP_RECVIF:
1435                 case IP_RECVTTL:
1436                 case IP_FAITH:
1437                         error = sooptcopyin(sopt, &optval, sizeof optval,
1438                                             sizeof optval);
1439                         if (error)
1440                                 break;
1441
1442                         switch (sopt->sopt_name) {
1443                         case IP_TOS:
1444                                 inp->inp_ip_tos = optval;
1445                                 break;
1446
1447                         case IP_TTL:
1448                                 inp->inp_ip_ttl = optval;
1449                                 break;
1450                         case IP_MINTTL:
1451                                 if (optval > 0 && optval <= MAXTTL)
1452                                         inp->inp_ip_minttl = optval;
1453                                 else
1454                                         error = EINVAL;
1455                                 break;
1456 #define OPTSET(bit) \
1457         if (optval) \
1458                 inp->inp_flags |= bit; \
1459         else \
1460                 inp->inp_flags &= ~bit;
1461
1462                         case IP_RECVOPTS:
1463                                 OPTSET(INP_RECVOPTS);
1464                                 break;
1465
1466                         case IP_RECVRETOPTS:
1467                                 OPTSET(INP_RECVRETOPTS);
1468                                 break;
1469
1470                         case IP_RECVDSTADDR:
1471                                 OPTSET(INP_RECVDSTADDR);
1472                                 break;
1473
1474                         case IP_RECVIF:
1475                                 OPTSET(INP_RECVIF);
1476                                 break;
1477
1478                         case IP_RECVTTL:
1479                                 OPTSET(INP_RECVTTL);
1480                                 break;
1481
1482                         case IP_FAITH:
1483                                 OPTSET(INP_FAITH);
1484                                 break;
1485                         }
1486                         break;
1487 #undef OPTSET
1488
1489                 case IP_MULTICAST_IF:
1490                 case IP_MULTICAST_VIF:
1491                 case IP_MULTICAST_TTL:
1492                 case IP_MULTICAST_LOOP:
1493                 case IP_ADD_MEMBERSHIP:
1494                 case IP_DROP_MEMBERSHIP:
1495                         error = ip_setmoptions(sopt, &inp->inp_moptions);
1496                         break;
1497
1498                 case IP_PORTRANGE:
1499                         error = sooptcopyin(sopt, &optval, sizeof optval,
1500                                             sizeof optval);
1501                         if (error)
1502                                 break;
1503
1504                         switch (optval) {
1505                         case IP_PORTRANGE_DEFAULT:
1506                                 inp->inp_flags &= ~(INP_LOWPORT);
1507                                 inp->inp_flags &= ~(INP_HIGHPORT);
1508                                 break;
1509
1510                         case IP_PORTRANGE_HIGH:
1511                                 inp->inp_flags &= ~(INP_LOWPORT);
1512                                 inp->inp_flags |= INP_HIGHPORT;
1513                                 break;
1514
1515                         case IP_PORTRANGE_LOW:
1516                                 inp->inp_flags &= ~(INP_HIGHPORT);
1517                                 inp->inp_flags |= INP_LOWPORT;
1518                                 break;
1519
1520                         default:
1521                                 error = EINVAL;
1522                                 break;
1523                         }
1524                         break;
1525
1526 #if defined(IPSEC) || defined(FAST_IPSEC)
1527                 case IP_IPSEC_POLICY:
1528                 {
1529                         caddr_t req;
1530                         size_t len = 0;
1531                         int priv;
1532                         struct mbuf *m;
1533                         int optname;
1534
1535                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1536                                 break;
1537                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1538                                 break;
1539                         priv = (sopt->sopt_td != NULL &&
1540                                 suser(sopt->sopt_td) != 0) ? 0 : 1;
1541                         req = mtod(m, caddr_t);
1542                         len = m->m_len;
1543                         optname = sopt->sopt_name;
1544                         error = ipsec4_set_policy(inp, optname, req, len, priv);
1545                         m_freem(m);
1546                         break;
1547                 }
1548 #endif /*IPSEC*/
1549
1550                 default:
1551                         error = ENOPROTOOPT;
1552                         break;
1553                 }
1554                 break;
1555
1556         case SOPT_GET:
1557                 switch (sopt->sopt_name) {
1558                 case IP_OPTIONS:
1559                 case IP_RETOPTS:
1560                         if (inp->inp_options)
1561                                 error = sooptcopyout(sopt,
1562                                                      mtod(inp->inp_options,
1563                                                           char *),
1564                                                      inp->inp_options->m_len);
1565                         else
1566                                 sopt->sopt_valsize = 0;
1567                         break;
1568
1569                 case IP_TOS:
1570                 case IP_TTL:
1571                 case IP_MINTTL:
1572                 case IP_RECVOPTS:
1573                 case IP_RECVRETOPTS:
1574                 case IP_RECVDSTADDR:
1575                 case IP_RECVTTL:
1576                 case IP_RECVIF:
1577                 case IP_PORTRANGE:
1578                 case IP_FAITH:
1579                         switch (sopt->sopt_name) {
1580
1581                         case IP_TOS:
1582                                 optval = inp->inp_ip_tos;
1583                                 break;
1584
1585                         case IP_TTL:
1586                                 optval = inp->inp_ip_ttl;
1587                                 break;
1588                         case IP_MINTTL:
1589                                 optval = inp->inp_ip_minttl;
1590                                 break;
1591
1592 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1593
1594                         case IP_RECVOPTS:
1595                                 optval = OPTBIT(INP_RECVOPTS);
1596                                 break;
1597
1598                         case IP_RECVRETOPTS:
1599                                 optval = OPTBIT(INP_RECVRETOPTS);
1600                                 break;
1601
1602                         case IP_RECVDSTADDR:
1603                                 optval = OPTBIT(INP_RECVDSTADDR);
1604                                 break;
1605
1606                         case IP_RECVTTL:
1607                                 optval = OPTBIT(INP_RECVTTL);
1608                                 break;
1609
1610                         case IP_RECVIF:
1611                                 optval = OPTBIT(INP_RECVIF);
1612                                 break;
1613
1614                         case IP_PORTRANGE:
1615                                 if (inp->inp_flags & INP_HIGHPORT)
1616                                         optval = IP_PORTRANGE_HIGH;
1617                                 else if (inp->inp_flags & INP_LOWPORT)
1618                                         optval = IP_PORTRANGE_LOW;
1619                                 else
1620                                         optval = 0;
1621                                 break;
1622
1623                         case IP_FAITH:
1624                                 optval = OPTBIT(INP_FAITH);
1625                                 break;
1626                         }
1627                         error = sooptcopyout(sopt, &optval, sizeof optval);
1628                         break;
1629
1630                 case IP_MULTICAST_IF:
1631                 case IP_MULTICAST_VIF:
1632                 case IP_MULTICAST_TTL:
1633                 case IP_MULTICAST_LOOP:
1634                 case IP_ADD_MEMBERSHIP:
1635                 case IP_DROP_MEMBERSHIP:
1636                         error = ip_getmoptions(sopt, inp->inp_moptions);
1637                         break;
1638
1639 #if defined(IPSEC) || defined(FAST_IPSEC)
1640                 case IP_IPSEC_POLICY:
1641                 {
1642                         struct mbuf *m = NULL;
1643                         caddr_t req = NULL;
1644                         size_t len = 0;
1645
1646                         if (m != NULL) {
1647                                 req = mtod(m, caddr_t);
1648                                 len = m->m_len;
1649                         }
1650                         error = ipsec4_get_policy(so->so_pcb, req, len, &m);
1651                         if (error == 0)
1652                                 error = soopt_mcopyout(sopt, m); /* XXX */
1653                         if (error == 0)
1654                                 m_freem(m);
1655                         break;
1656                 }
1657 #endif /*IPSEC*/
1658
1659                 default:
1660                         error = ENOPROTOOPT;
1661                         break;
1662                 }
1663                 break;
1664         }
1665         return (error);
1666 }
1667
1668 /*
1669  * Set up IP options in pcb for insertion in output packets.
1670  * Store in mbuf with pointer in pcbopt, adding pseudo-option
1671  * with destination address if source routed.
1672  */
1673 static int
1674 ip_pcbopts(int optname, struct mbuf **pcbopt, struct mbuf *m)
1675 {
1676         int cnt, optlen;
1677         u_char *cp;
1678         u_char opt;
1679
1680         /* turn off any old options */
1681         if (*pcbopt)
1682                 m_free(*pcbopt);
1683         *pcbopt = 0;
1684         if (m == NULL || m->m_len == 0) {
1685                 /*
1686                  * Only turning off any previous options.
1687                  */
1688                 if (m != NULL)
1689                         m_free(m);
1690                 return (0);
1691         }
1692
1693         if (m->m_len % sizeof(int32_t))
1694                 goto bad;
1695         /*
1696          * IP first-hop destination address will be stored before
1697          * actual options; move other options back
1698          * and clear it when none present.
1699          */
1700         if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN])
1701                 goto bad;
1702         cnt = m->m_len;
1703         m->m_len += sizeof(struct in_addr);
1704         cp = mtod(m, u_char *) + sizeof(struct in_addr);
1705         ovbcopy(mtod(m, caddr_t), cp, cnt);
1706         bzero(mtod(m, caddr_t), sizeof(struct in_addr));
1707
1708         for (; cnt > 0; cnt -= optlen, cp += optlen) {
1709                 opt = cp[IPOPT_OPTVAL];
1710                 if (opt == IPOPT_EOL)
1711                         break;
1712                 if (opt == IPOPT_NOP)
1713                         optlen = 1;
1714                 else {
1715                         if (cnt < IPOPT_OLEN + sizeof *cp)
1716                                 goto bad;
1717                         optlen = cp[IPOPT_OLEN];
1718                         if (optlen < IPOPT_OLEN + sizeof *cp || optlen > cnt)
1719                                 goto bad;
1720                 }
1721                 switch (opt) {
1722
1723                 default:
1724                         break;
1725
1726                 case IPOPT_LSRR:
1727                 case IPOPT_SSRR:
1728                         /*
1729                          * user process specifies route as:
1730                          *      ->A->B->C->D
1731                          * D must be our final destination (but we can't
1732                          * check that since we may not have connected yet).
1733                          * A is first hop destination, which doesn't appear in
1734                          * actual IP option, but is stored before the options.
1735                          */
1736                         if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr))
1737                                 goto bad;
1738                         m->m_len -= sizeof(struct in_addr);
1739                         cnt -= sizeof(struct in_addr);
1740                         optlen -= sizeof(struct in_addr);
1741                         cp[IPOPT_OLEN] = optlen;
1742                         /*
1743                          * Move first hop before start of options.
1744                          */
1745                         bcopy(&cp[IPOPT_OFFSET+1], mtod(m, caddr_t),
1746                               sizeof(struct in_addr));
1747                         /*
1748                          * Then copy rest of options back
1749                          * to close up the deleted entry.
1750                          */
1751                         ovbcopy(&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr),
1752                                 &cp[IPOPT_OFFSET+1],
1753                                 cnt - (IPOPT_MINOFF - 1));
1754                         break;
1755                 }
1756         }
1757         if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr))
1758                 goto bad;
1759         *pcbopt = m;
1760         return (0);
1761
1762 bad:
1763         m_free(m);
1764         return (EINVAL);
1765 }
1766
1767 /*
1768  * XXX
1769  * The whole multicast option thing needs to be re-thought.
1770  * Several of these options are equally applicable to non-multicast
1771  * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
1772  * standard option (IP_TTL).
1773  */
1774
1775 /*
1776  * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
1777  */
1778 static struct ifnet *
1779 ip_multicast_if(struct in_addr *a, int *ifindexp)
1780 {
1781         int ifindex;
1782         struct ifnet *ifp;
1783
1784         if (ifindexp)
1785                 *ifindexp = 0;
1786         if (ntohl(a->s_addr) >> 24 == 0) {
1787                 ifindex = ntohl(a->s_addr) & 0xffffff;
1788                 if (ifindex < 0 || if_index < ifindex)
1789                         return NULL;
1790                 ifp = ifindex2ifnet[ifindex];
1791                 if (ifindexp)
1792                         *ifindexp = ifindex;
1793         } else {
1794                 INADDR_TO_IFP(*a, ifp);
1795         }
1796         return ifp;
1797 }
1798
1799 /*
1800  * Set the IP multicast options in response to user setsockopt().
1801  */
1802 static int
1803 ip_setmoptions(struct sockopt *sopt, struct ip_moptions **imop)
1804 {
1805         int error = 0;
1806         int i;
1807         struct in_addr addr;
1808         struct ip_mreq mreq;
1809         struct ifnet *ifp;
1810         struct ip_moptions *imo = *imop;
1811         int ifindex;
1812
1813         if (imo == NULL) {
1814                 /*
1815                  * No multicast option buffer attached to the pcb;
1816                  * allocate one and initialize to default values.
1817                  */
1818                 imo = kmalloc(sizeof *imo, M_IPMOPTS, M_WAITOK);
1819
1820                 *imop = imo;
1821                 imo->imo_multicast_ifp = NULL;
1822                 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1823                 imo->imo_multicast_vif = -1;
1824                 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1825                 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
1826                 imo->imo_num_memberships = 0;
1827         }
1828
1829         switch (sopt->sopt_name) {
1830         /* store an index number for the vif you wanna use in the send */
1831         case IP_MULTICAST_VIF:
1832                 if (legal_vif_num == 0) {
1833                         error = EOPNOTSUPP;
1834                         break;
1835                 }
1836                 error = sooptcopyin(sopt, &i, sizeof i, sizeof i);
1837                 if (error)
1838                         break;
1839                 if (!legal_vif_num(i) && (i != -1)) {
1840                         error = EINVAL;
1841                         break;
1842                 }
1843                 imo->imo_multicast_vif = i;
1844                 break;
1845
1846         case IP_MULTICAST_IF:
1847                 /*
1848                  * Select the interface for outgoing multicast packets.
1849                  */
1850                 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr);
1851                 if (error)
1852                         break;
1853                 /*
1854                  * INADDR_ANY is used to remove a previous selection.
1855                  * When no interface is selected, a default one is
1856                  * chosen every time a multicast packet is sent.
1857                  */
1858                 if (addr.s_addr == INADDR_ANY) {
1859                         imo->imo_multicast_ifp = NULL;
1860                         break;
1861                 }
1862                 /*
1863                  * The selected interface is identified by its local
1864                  * IP address.  Find the interface and confirm that
1865                  * it supports multicasting.
1866                  */
1867                 crit_enter();
1868                 ifp = ip_multicast_if(&addr, &ifindex);
1869                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1870                         crit_exit();
1871                         error = EADDRNOTAVAIL;
1872                         break;
1873                 }
1874                 imo->imo_multicast_ifp = ifp;
1875                 if (ifindex)
1876                         imo->imo_multicast_addr = addr;
1877                 else
1878                         imo->imo_multicast_addr.s_addr = INADDR_ANY;
1879                 crit_exit();
1880                 break;
1881
1882         case IP_MULTICAST_TTL:
1883                 /*
1884                  * Set the IP time-to-live for outgoing multicast packets.
1885                  * The original multicast API required a char argument,
1886                  * which is inconsistent with the rest of the socket API.
1887                  * We allow either a char or an int.
1888                  */
1889                 if (sopt->sopt_valsize == 1) {
1890                         u_char ttl;
1891                         error = sooptcopyin(sopt, &ttl, 1, 1);
1892                         if (error)
1893                                 break;
1894                         imo->imo_multicast_ttl = ttl;
1895                 } else {
1896                         u_int ttl;
1897                         error = sooptcopyin(sopt, &ttl, sizeof ttl, sizeof ttl);
1898                         if (error)
1899                                 break;
1900                         if (ttl > 255)
1901                                 error = EINVAL;
1902                         else
1903                                 imo->imo_multicast_ttl = ttl;
1904                 }
1905                 break;
1906
1907         case IP_MULTICAST_LOOP:
1908                 /*
1909                  * Set the loopback flag for outgoing multicast packets.
1910                  * Must be zero or one.  The original multicast API required a
1911                  * char argument, which is inconsistent with the rest
1912                  * of the socket API.  We allow either a char or an int.
1913                  */
1914                 if (sopt->sopt_valsize == 1) {
1915                         u_char loop;
1916
1917                         error = sooptcopyin(sopt, &loop, 1, 1);
1918                         if (error)
1919                                 break;
1920                         imo->imo_multicast_loop = !!loop;
1921                 } else {
1922                         u_int loop;
1923
1924                         error = sooptcopyin(sopt, &loop, sizeof loop,
1925                                             sizeof loop);
1926                         if (error)
1927                                 break;
1928                         imo->imo_multicast_loop = !!loop;
1929                 }
1930                 break;
1931
1932         case IP_ADD_MEMBERSHIP:
1933                 /*
1934                  * Add a multicast group membership.
1935                  * Group must be a valid IP multicast address.
1936                  */
1937                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
1938                 if (error)
1939                         break;
1940
1941                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
1942                         error = EINVAL;
1943                         break;
1944                 }
1945                 crit_enter();
1946                 /*
1947                  * If no interface address was provided, use the interface of
1948                  * the route to the given multicast address.
1949                  */
1950                 if (mreq.imr_interface.s_addr == INADDR_ANY) {
1951                         struct sockaddr_in dst;
1952                         struct rtentry *rt;
1953
1954                         bzero(&dst, sizeof(struct sockaddr_in));
1955                         dst.sin_len = sizeof(struct sockaddr_in);
1956                         dst.sin_family = AF_INET;
1957                         dst.sin_addr = mreq.imr_multiaddr;
1958                         rt = rtlookup((struct sockaddr *)&dst);
1959                         if (rt == NULL) {
1960                                 error = EADDRNOTAVAIL;
1961                                 crit_exit();
1962                                 break;
1963                         }
1964                         --rt->rt_refcnt;
1965                         ifp = rt->rt_ifp;
1966                 } else {
1967                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
1968                 }
1969
1970                 /*
1971                  * See if we found an interface, and confirm that it
1972                  * supports multicast.
1973                  */
1974                 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
1975                         error = EADDRNOTAVAIL;
1976                         crit_exit();
1977                         break;
1978                 }
1979                 /*
1980                  * See if the membership already exists or if all the
1981                  * membership slots are full.
1982                  */
1983                 for (i = 0; i < imo->imo_num_memberships; ++i) {
1984                         if (imo->imo_membership[i]->inm_ifp == ifp &&
1985                             imo->imo_membership[i]->inm_addr.s_addr
1986                                                 == mreq.imr_multiaddr.s_addr)
1987                                 break;
1988                 }
1989                 if (i < imo->imo_num_memberships) {
1990                         error = EADDRINUSE;
1991                         crit_exit();
1992                         break;
1993                 }
1994                 if (i == IP_MAX_MEMBERSHIPS) {
1995                         error = ETOOMANYREFS;
1996                         crit_exit();
1997                         break;
1998                 }
1999                 /*
2000                  * Everything looks good; add a new record to the multicast
2001                  * address list for the given interface.
2002                  */
2003                 if ((imo->imo_membership[i] =
2004                      in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) {
2005                         error = ENOBUFS;
2006                         crit_exit();
2007                         break;
2008                 }
2009                 ++imo->imo_num_memberships;
2010                 crit_exit();
2011                 break;
2012
2013         case IP_DROP_MEMBERSHIP:
2014                 /*
2015                  * Drop a multicast group membership.
2016                  * Group must be a valid IP multicast address.
2017                  */
2018                 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq);
2019                 if (error)
2020                         break;
2021
2022                 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) {
2023                         error = EINVAL;
2024                         break;
2025                 }
2026
2027                 crit_enter();
2028                 /*
2029                  * If an interface address was specified, get a pointer
2030                  * to its ifnet structure.
2031                  */
2032                 if (mreq.imr_interface.s_addr == INADDR_ANY)
2033                         ifp = NULL;
2034                 else {
2035                         ifp = ip_multicast_if(&mreq.imr_interface, NULL);
2036                         if (ifp == NULL) {
2037                                 error = EADDRNOTAVAIL;
2038                                 crit_exit();
2039                                 break;
2040                         }
2041                 }
2042                 /*
2043                  * Find the membership in the membership array.
2044                  */
2045                 for (i = 0; i < imo->imo_num_memberships; ++i) {
2046                         if ((ifp == NULL ||
2047                              imo->imo_membership[i]->inm_ifp == ifp) &&
2048                             imo->imo_membership[i]->inm_addr.s_addr ==
2049                             mreq.imr_multiaddr.s_addr)
2050                                 break;
2051                 }
2052                 if (i == imo->imo_num_memberships) {
2053                         error = EADDRNOTAVAIL;
2054                         crit_exit();
2055                         break;
2056                 }
2057                 /*
2058                  * Give up the multicast address record to which the
2059                  * membership points.
2060                  */
2061                 in_delmulti(imo->imo_membership[i]);
2062                 /*
2063                  * Remove the gap in the membership array.
2064                  */
2065                 for (++i; i < imo->imo_num_memberships; ++i)
2066                         imo->imo_membership[i-1] = imo->imo_membership[i];
2067                 --imo->imo_num_memberships;
2068                 crit_exit();
2069                 break;
2070
2071         default:
2072                 error = EOPNOTSUPP;
2073                 break;
2074         }
2075
2076         /*
2077          * If all options have default values, no need to keep the mbuf.
2078          */
2079         if (imo->imo_multicast_ifp == NULL &&
2080             imo->imo_multicast_vif == -1 &&
2081             imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL &&
2082             imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP &&
2083             imo->imo_num_memberships == 0) {
2084                 kfree(*imop, M_IPMOPTS);
2085                 *imop = NULL;
2086         }
2087
2088         return (error);
2089 }
2090
2091 /*
2092  * Return the IP multicast options in response to user getsockopt().
2093  */
2094 static int
2095 ip_getmoptions(struct sockopt *sopt, struct ip_moptions *imo)
2096 {
2097         struct in_addr addr;
2098         struct in_ifaddr *ia;
2099         int error, optval;
2100         u_char coptval;
2101
2102         error = 0;
2103         switch (sopt->sopt_name) {
2104         case IP_MULTICAST_VIF:
2105                 if (imo != NULL)
2106                         optval = imo->imo_multicast_vif;
2107                 else
2108                         optval = -1;
2109                 error = sooptcopyout(sopt, &optval, sizeof optval);
2110                 break;
2111
2112         case IP_MULTICAST_IF:
2113                 if (imo == NULL || imo->imo_multicast_ifp == NULL)
2114                         addr.s_addr = INADDR_ANY;
2115                 else if (imo->imo_multicast_addr.s_addr) {
2116                         /* return the value user has set */
2117                         addr = imo->imo_multicast_addr;
2118                 } else {
2119                         IFP_TO_IA(imo->imo_multicast_ifp, ia);
2120                         addr.s_addr = (ia == NULL) ? INADDR_ANY
2121                                 : IA_SIN(ia)->sin_addr.s_addr;
2122                 }
2123                 error = sooptcopyout(sopt, &addr, sizeof addr);
2124                 break;
2125
2126         case IP_MULTICAST_TTL:
2127                 if (imo == NULL)
2128                         optval = coptval = IP_DEFAULT_MULTICAST_TTL;
2129                 else
2130                         optval = coptval = imo->imo_multicast_ttl;
2131                 if (sopt->sopt_valsize == 1)
2132                         error = sooptcopyout(sopt, &coptval, 1);
2133                 else
2134                         error = sooptcopyout(sopt, &optval, sizeof optval);
2135                 break;
2136
2137         case IP_MULTICAST_LOOP:
2138                 if (imo == NULL)
2139                         optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
2140                 else
2141                         optval = coptval = imo->imo_multicast_loop;
2142                 if (sopt->sopt_valsize == 1)
2143                         error = sooptcopyout(sopt, &coptval, 1);
2144                 else
2145                         error = sooptcopyout(sopt, &optval, sizeof optval);
2146                 break;
2147
2148         default:
2149                 error = ENOPROTOOPT;
2150                 break;
2151         }
2152         return (error);
2153 }
2154
2155 /*
2156  * Discard the IP multicast options.
2157  */
2158 void
2159 ip_freemoptions(struct ip_moptions *imo)
2160 {
2161         int i;
2162
2163         if (imo != NULL) {
2164                 for (i = 0; i < imo->imo_num_memberships; ++i)
2165                         in_delmulti(imo->imo_membership[i]);
2166                 kfree(imo, M_IPMOPTS);
2167         }
2168 }
2169
2170 /*
2171  * Routine called from ip_output() to loop back a copy of an IP multicast
2172  * packet to the input queue of a specified interface.  Note that this
2173  * calls the output routine of the loopback "driver", but with an interface
2174  * pointer that might NOT be a loopback interface -- evil, but easier than
2175  * replicating that code here.
2176  */
2177 static void
2178 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst,
2179              int hlen)
2180 {
2181         struct ip *ip;
2182         struct mbuf *copym;
2183
2184         copym = m_copypacket(m, MB_DONTWAIT);
2185         if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen))
2186                 copym = m_pullup(copym, hlen);
2187         if (copym != NULL) {
2188                 /*
2189                  * if the checksum hasn't been computed, mark it as valid
2190                  */
2191                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
2192                         in_delayed_cksum(copym);
2193                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
2194                         copym->m_pkthdr.csum_flags |=
2195                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2196                         copym->m_pkthdr.csum_data = 0xffff;
2197                 }
2198                 /*
2199                  * We don't bother to fragment if the IP length is greater
2200                  * than the interface's MTU.  Can this possibly matter?
2201                  */
2202                 ip = mtod(copym, struct ip *);
2203                 ip->ip_len = htons(ip->ip_len);
2204                 ip->ip_off = htons(ip->ip_off);
2205                 ip->ip_sum = 0;
2206                 if (ip->ip_vhl == IP_VHL_BORING) {
2207                         ip->ip_sum = in_cksum_hdr(ip);
2208                 } else {
2209                         ip->ip_sum = in_cksum(copym, hlen);
2210                 }
2211                 /*
2212                  * NB:
2213                  * It's not clear whether there are any lingering
2214                  * reentrancy problems in other areas which might
2215                  * be exposed by using ip_input directly (in
2216                  * particular, everything which modifies the packet
2217                  * in-place).  Yet another option is using the
2218                  * protosw directly to deliver the looped back
2219                  * packet.  For the moment, we'll err on the side
2220                  * of safety by using if_simloop().
2221                  */
2222 #if 1 /* XXX */
2223                 if (dst->sin_family != AF_INET) {
2224                         kprintf("ip_mloopback: bad address family %d\n",
2225                                                 dst->sin_family);
2226                         dst->sin_family = AF_INET;
2227                 }
2228 #endif
2229
2230 #ifdef notdef
2231                 copym->m_pkthdr.rcvif = ifp;
2232                 ip_input(copym);
2233 #else
2234                 if_simloop(ifp, copym, dst->sin_family, 0);
2235 #endif
2236         }
2237 }